journal.o \
journal_io.o \
journal_reclaim.o \
+ journal_sb.o \
journal_seq_blacklist.o \
keylist.o \
migrate.o \
__le32 type;
};
-#define BCH_SB_FIELDS() \
- x(journal, 0) \
- x(members, 1) \
- x(crypt, 2) \
- x(replicas_v0, 3) \
- x(quota, 4) \
- x(disk_groups, 5) \
- x(clean, 6) \
- x(replicas, 7) \
- x(journal_seq_blacklist, 8)
+#define BCH_SB_FIELDS() \
+ x(journal, 0) \
+ x(members, 1) \
+ x(crypt, 2) \
+ x(replicas_v0, 3) \
+ x(quota, 4) \
+ x(disk_groups, 5) \
+ x(clean, 6) \
+ x(replicas, 7) \
+ x(journal_seq_blacklist, 8) \
+ x(journal_v2, 9)
enum bch_sb_field_type {
#define x(f, nr) BCH_SB_FIELD_##f = nr,
BCH_SB_FIELD_NR
};
+/*
+ * Most superblock fields are replicated in all device's superblocks - a few are
+ * not:
+ */
+#define BCH_SINGLE_DEVICE_SB_FIELDS \
+ ((1U << BCH_SB_FIELD_journal)| \
+ (1U << BCH_SB_FIELD_journal_v2))
+
/* BCH_SB_FIELD_journal: */
struct bch_sb_field_journal {
__le64 buckets[0];
};
+struct bch_sb_field_journal_v2 {
+ struct bch_sb_field field;
+
+ struct bch_sb_field_journal_v2_entry {
+ __le64 start;
+ __le64 nr;
+ } d[0];
+};
+
/* BCH_SB_FIELD_members: */
#define BCH_MIN_NR_NBUCKETS (1 << 6)
#include "journal.h"
#include "journal_io.h"
#include "journal_reclaim.h"
+#include "journal_sb.h"
#include "journal_seq_blacklist.h"
-#include "super-io.h"
#include "trace.h"
#define x(n) #n,
{
struct bch_fs *c = ca->fs;
struct journal_device *ja = &ca->journal;
- struct bch_sb_field_journal *journal_buckets;
u64 *new_bucket_seq = NULL, *new_buckets = NULL;
+ struct open_bucket **ob = NULL;
+ long *bu = NULL;
+ unsigned i, nr_got = 0, nr_want = nr - ja->nr;
+ unsigned old_nr = ja->nr;
+ unsigned old_discard_idx = ja->discard_idx;
+ unsigned old_dirty_idx_ondisk = ja->dirty_idx_ondisk;
+ unsigned old_dirty_idx = ja->dirty_idx;
+ unsigned old_cur_idx = ja->cur_idx;
int ret = 0;
- /* don't handle reducing nr of buckets yet: */
- if (nr <= ja->nr)
- return 0;
+ if (c) {
+ bch2_journal_flush_all_pins(&c->journal);
+ bch2_journal_block(&c->journal);
+ }
+ bu = kzalloc(nr_want * sizeof(*bu), GFP_KERNEL);
+ ob = kzalloc(nr_want * sizeof(*ob), GFP_KERNEL);
new_buckets = kzalloc(nr * sizeof(u64), GFP_KERNEL);
new_bucket_seq = kzalloc(nr * sizeof(u64), GFP_KERNEL);
- if (!new_buckets || !new_bucket_seq) {
+ if (!bu || !ob || !new_buckets || !new_bucket_seq) {
ret = -ENOMEM;
- goto err;
+ goto err_unblock;
}
- journal_buckets = bch2_sb_resize_journal(&ca->disk_sb,
- nr + sizeof(*journal_buckets) / sizeof(u64));
- if (!journal_buckets) {
- ret = -ENOSPC;
- goto err;
+ for (nr_got = 0; nr_got < nr_want; nr_got++) {
+ if (new_fs) {
+ bu[nr_got] = bch2_bucket_alloc_new_fs(ca);
+ if (bu[nr_got] < 0) {
+ ret = -ENOSPC;
+ break;
+ }
+ } else {
+ rcu_read_lock();
+ ob[nr_got] = bch2_bucket_alloc(c, ca, RESERVE_none,
+ false, cl);
+ rcu_read_unlock();
+ if (IS_ERR(ob[nr_got])) {
+ ret = cl ? -EAGAIN : -ENOSPC;
+ break;
+ }
+
+ bu[nr_got] = ob[nr_got]->bucket;
+ }
}
+ if (!nr_got)
+ goto err_unblock;
+
/*
* We may be called from the device add path, before the new device has
* actually been added to the running filesystem:
swap(new_buckets, ja->buckets);
swap(new_bucket_seq, ja->bucket_seq);
- if (!new_fs)
- spin_unlock(&c->journal.lock);
-
- while (ja->nr < nr) {
- struct open_bucket *ob = NULL;
- unsigned pos;
- long b;
-
- if (new_fs) {
- b = bch2_bucket_alloc_new_fs(ca);
- if (b < 0) {
- ret = -ENOSPC;
- goto err;
- }
- } else {
- rcu_read_lock();
- ob = bch2_bucket_alloc(c, ca, RESERVE_none,
- false, cl);
- rcu_read_unlock();
- if (IS_ERR(ob)) {
- ret = cl ? -EAGAIN : -ENOSPC;
- goto err;
- }
-
- b = ob->bucket;
- }
+ for (i = 0; i < nr_got; i++) {
+ unsigned pos = ja->discard_idx ?: ja->nr;
+ long b = bu[i];
- if (c)
- spin_lock(&c->journal.lock);
-
- /*
- * XXX
- * For resize at runtime, we should be writing the new
- * superblock before inserting into the journal array
- */
-
- pos = ja->discard_idx ?: ja->nr;
__array_insert_item(ja->buckets, ja->nr, pos);
__array_insert_item(ja->bucket_seq, ja->nr, pos);
- __array_insert_item(journal_buckets->buckets, ja->nr, pos);
ja->nr++;
ja->buckets[pos] = b;
ja->bucket_seq[pos] = 0;
- journal_buckets->buckets[pos] = cpu_to_le64(b);
if (pos <= ja->discard_idx)
ja->discard_idx = (ja->discard_idx + 1) % ja->nr;
ja->dirty_idx = (ja->dirty_idx + 1) % ja->nr;
if (pos <= ja->cur_idx)
ja->cur_idx = (ja->cur_idx + 1) % ja->nr;
+ }
- if (c)
- spin_unlock(&c->journal.lock);
+ ret = bch2_journal_buckets_to_sb(c, ca);
+ if (ret) {
+ /* Revert: */
+ swap(new_buckets, ja->buckets);
+ swap(new_bucket_seq, ja->bucket_seq);
+ ja->nr = old_nr;
+ ja->discard_idx = old_discard_idx;
+ ja->dirty_idx_ondisk = old_dirty_idx_ondisk;
+ ja->dirty_idx = old_dirty_idx;
+ ja->cur_idx = old_cur_idx;
+ }
+
+ if (!new_fs)
+ spin_unlock(&c->journal.lock);
- if (!new_fs) {
+ if (c)
+ bch2_journal_unblock(&c->journal);
+
+ if (ret)
+ goto err;
+
+ if (!new_fs) {
+ for (i = 0; i < nr_got; i++) {
ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_NOFAIL,
bch2_trans_mark_metadata_bucket(&trans, ca,
- b, BCH_DATA_journal,
+ bu[i], BCH_DATA_journal,
ca->mi.bucket_size));
-
- bch2_open_bucket_put(c, ob);
-
- if (ret)
+ if (ret) {
+ bch2_fs_inconsistent(c, "error marking new journal buckets: %i", ret);
goto err;
+ }
}
}
err:
- bch2_sb_resize_journal(&ca->disk_sb,
- ja->nr + sizeof(*journal_buckets) / sizeof(u64));
+ if (ob && !new_fs)
+ for (i = 0; i < nr_got; i++)
+ bch2_open_bucket_put(c, ob[i]);
+
kfree(new_bucket_seq);
kfree(new_buckets);
+ kfree(ob);
+ kfree(bu);
return ret;
+err_unblock:
+ if (c)
+ bch2_journal_unblock(&c->journal);
+ goto err;
}
/*
struct journal_device *ja = &ca->journal;
struct closure cl;
unsigned current_nr;
- int ret;
+ int ret = 0;
+
+ /* don't handle reducing nr of buckets yet: */
+ if (nr < ja->nr)
+ return 0;
closure_init_stack(&cl);
- do {
+ while (ja->nr != nr && (ret == 0 || ret == -EAGAIN)) {
struct disk_reservation disk_res = { 0, 0 };
closure_sync(&cl);
if (ja->nr != current_nr)
bch2_write_super(c);
mutex_unlock(&c->sb_lock);
- } while (ret == -EAGAIN);
+ }
return ret;
}
int bch2_dev_journal_alloc(struct bch_dev *ca)
{
unsigned nr;
+ int ret;
if (dynamic_fault("bcachefs:add:journal_alloc"))
return -ENOMEM;
min(1 << 13,
(1 << 24) / ca->mi.bucket_size));
- return __bch2_set_nr_journal_buckets(ca, nr, true, NULL);
+ if (ca->fs)
+ mutex_lock(&ca->fs->sb_lock);
+
+ ret = __bch2_set_nr_journal_buckets(ca, nr, true, NULL);
+
+ if (ca->fs)
+ mutex_unlock(&ca->fs->sb_lock);
+
+ return ret;
}
/* startup/shutdown: */
struct journal_device *ja = &ca->journal;
struct bch_sb_field_journal *journal_buckets =
bch2_sb_get_journal(sb);
+ struct bch_sb_field_journal_v2 *journal_buckets_v2 =
+ bch2_sb_get_journal_v2(sb);
unsigned i, nr_bvecs;
- ja->nr = bch2_nr_journal_buckets(journal_buckets);
+ ja->nr = 0;
+
+ if (journal_buckets_v2) {
+ unsigned nr = bch2_sb_field_journal_v2_nr_entries(journal_buckets_v2);
+
+ for (i = 0; i < nr; i++)
+ ja->nr += le64_to_cpu(journal_buckets_v2->d[i].nr);
+ } else if (journal_buckets) {
+ ja->nr = bch2_nr_journal_buckets(journal_buckets);
+ }
ja->bucket_seq = kcalloc(ja->nr, sizeof(u64), GFP_KERNEL);
if (!ja->bucket_seq)
if (!ja->buckets)
return -ENOMEM;
- for (i = 0; i < ja->nr; i++)
- ja->buckets[i] = le64_to_cpu(journal_buckets->buckets[i]);
+ if (journal_buckets_v2) {
+ unsigned nr = bch2_sb_field_journal_v2_nr_entries(journal_buckets_v2);
+ unsigned j, dst = 0;
+
+ for (i = 0; i < nr; i++)
+ for (j = 0; j < le64_to_cpu(journal_buckets_v2->d[i].nr); j++)
+ ja->buckets[dst++] =
+ le64_to_cpu(journal_buckets_v2->d[i].start) + j;
+ } else if (journal_buckets) {
+ for (i = 0; i < ja->nr; i++)
+ ja->buckets[i] = le64_to_cpu(journal_buckets->buckets[i]);
+ }
return 0;
}
--- /dev/null
+// SPDX-License-Identifier: GPL-2.0
+
+#include "bcachefs.h"
+#include "journal_sb.h"
+#include "darray.h"
+
+#include <linux/sort.h>
+
+/* BCH_SB_FIELD_journal: */
+
+static int u64_cmp(const void *_l, const void *_r)
+{
+ const u64 *l = _l;
+ const u64 *r = _r;
+
+ return cmp_int(*l, *r);
+}
+
+static int bch2_sb_journal_validate(struct bch_sb *sb,
+ struct bch_sb_field *f,
+ struct printbuf *err)
+{
+ struct bch_sb_field_journal *journal = field_to_type(f, journal);
+ struct bch_member *m = bch2_sb_get_members(sb)->members + sb->dev_idx;
+ int ret = -EINVAL;
+ unsigned nr;
+ unsigned i;
+ u64 *b;
+
+ nr = bch2_nr_journal_buckets(journal);
+ if (!nr)
+ return 0;
+
+ b = kmalloc_array(sizeof(u64), nr, GFP_KERNEL);
+ if (!b)
+ return -ENOMEM;
+
+ for (i = 0; i < nr; i++)
+ b[i] = le64_to_cpu(journal->buckets[i]);
+
+ sort(b, nr, sizeof(u64), u64_cmp, NULL);
+
+ if (!b[0]) {
+ pr_buf(err, "journal bucket at sector 0");
+ goto err;
+ }
+
+ if (b[0] < le16_to_cpu(m->first_bucket)) {
+ pr_buf(err, "journal bucket %llu before first bucket %u",
+ b[0], le16_to_cpu(m->first_bucket));
+ goto err;
+ }
+
+ if (b[nr - 1] >= le64_to_cpu(m->nbuckets)) {
+ pr_buf(err, "journal bucket %llu past end of device (nbuckets %llu)",
+ b[nr - 1], le64_to_cpu(m->nbuckets));
+ goto err;
+ }
+
+ for (i = 0; i + 1 < nr; i++)
+ if (b[i] == b[i + 1]) {
+ pr_buf(err, "duplicate journal buckets %llu", b[i]);
+ goto err;
+ }
+
+ ret = 0;
+err:
+ kfree(b);
+ return ret;
+}
+
+static void bch2_sb_journal_to_text(struct printbuf *out, struct bch_sb *sb,
+ struct bch_sb_field *f)
+{
+ struct bch_sb_field_journal *journal = field_to_type(f, journal);
+ unsigned i, nr = bch2_nr_journal_buckets(journal);
+
+ pr_buf(out, "Buckets: ");
+ for (i = 0; i < nr; i++)
+ pr_buf(out, " %llu", le64_to_cpu(journal->buckets[i]));
+ pr_newline(out);
+}
+
+const struct bch_sb_field_ops bch_sb_field_ops_journal = {
+ .validate = bch2_sb_journal_validate,
+ .to_text = bch2_sb_journal_to_text,
+};
+
+struct u64_range {
+ u64 start;
+ u64 end;
+};
+
+static int u64_range_cmp(const void *_l, const void *_r)
+{
+ const struct u64_range *l = _l;
+ const struct u64_range *r = _r;
+
+ return cmp_int(l->start, r->start);
+}
+
+static int bch2_sb_journal_v2_validate(struct bch_sb *sb,
+ struct bch_sb_field *f,
+ struct printbuf *err)
+{
+ struct bch_sb_field_journal_v2 *journal = field_to_type(f, journal_v2);
+ struct bch_member *m = bch2_sb_get_members(sb)->members + sb->dev_idx;
+ int ret = -EINVAL;
+ unsigned nr;
+ unsigned i;
+ struct u64_range *b;
+
+ nr = bch2_sb_field_journal_v2_nr_entries(journal);
+ if (!nr)
+ return 0;
+
+ b = kmalloc_array(sizeof(*b), nr, GFP_KERNEL);
+ if (!b)
+ return -ENOMEM;
+
+ for (i = 0; i < nr; i++) {
+ b[i].start = le64_to_cpu(journal->d[i].start);
+ b[i].end = b[i].start + le64_to_cpu(journal->d[i].nr);
+ }
+
+ sort(b, nr, sizeof(*b), u64_range_cmp, NULL);
+
+ if (!b[0].start) {
+ pr_buf(err, "journal bucket at sector 0");
+ goto err;
+ }
+
+ if (b[0].start < le16_to_cpu(m->first_bucket)) {
+ pr_buf(err, "journal bucket %llu before first bucket %u",
+ b[0].start, le16_to_cpu(m->first_bucket));
+ goto err;
+ }
+
+ if (b[nr - 1].end > le64_to_cpu(m->nbuckets)) {
+ pr_buf(err, "journal bucket %llu past end of device (nbuckets %llu)",
+ b[nr - 1].end - 1, le64_to_cpu(m->nbuckets));
+ goto err;
+ }
+
+ for (i = 0; i + 1 < nr; i++) {
+ if (b[i].end > b[i + 1].start) {
+ pr_buf(err, "duplicate journal buckets in ranges %llu-%llu, %llu-%llu",
+ b[i].start, b[i].end, b[i + 1].start, b[i + 1].end);
+ goto err;
+ }
+ }
+
+ ret = 0;
+err:
+ kfree(b);
+ return ret;
+}
+
+static void bch2_sb_journal_v2_to_text(struct printbuf *out, struct bch_sb *sb,
+ struct bch_sb_field *f)
+{
+ struct bch_sb_field_journal_v2 *journal = field_to_type(f, journal_v2);
+ unsigned i, nr = bch2_sb_field_journal_v2_nr_entries(journal);
+
+ pr_buf(out, "Buckets: ");
+ for (i = 0; i < nr; i++)
+ pr_buf(out, " %llu-%llu",
+ le64_to_cpu(journal->d[i].start),
+ le64_to_cpu(journal->d[i].start) + le64_to_cpu(journal->d[i].nr));
+ pr_newline(out);
+}
+
+const struct bch_sb_field_ops bch_sb_field_ops_journal_v2 = {
+ .validate = bch2_sb_journal_v2_validate,
+ .to_text = bch2_sb_journal_v2_to_text,
+};
+
+int bch2_journal_buckets_to_sb(struct bch_fs *c, struct bch_dev *ca)
+{
+ struct journal_device *ja = &ca->journal;
+ struct bch_sb_field_journal_v2 *j;
+ unsigned i, dst = 0, nr = 1;
+
+ if (c)
+ lockdep_assert_held(&c->sb_lock);
+
+ if (!ja->nr) {
+ bch2_sb_field_delete(&ca->disk_sb, BCH_SB_FIELD_journal);
+ bch2_sb_field_delete(&ca->disk_sb, BCH_SB_FIELD_journal_v2);
+ return 0;
+ }
+
+ for (i = 0; i + 1 < ja->nr; i++)
+ if (ja->buckets[i] + 1 != ja->buckets[i + 1])
+ nr++;
+
+ j = bch2_sb_resize_journal_v2(&ca->disk_sb,
+ (sizeof(*j) + sizeof(j->d[0]) * nr) / sizeof(u64));
+ if (!j)
+ return -ENOSPC;
+
+ bch2_sb_field_delete(&ca->disk_sb, BCH_SB_FIELD_journal);
+
+ j->d[dst].start = le64_to_cpu(ja->buckets[0]);
+ j->d[dst].nr = le64_to_cpu(1);
+
+ for (i = 1; i < ja->nr; i++) {
+ if (ja->buckets[i] == ja->buckets[i - 1] + 1) {
+ le64_add_cpu(&j->d[dst].nr, 1);
+ } else {
+ dst++;
+ j->d[dst].start = le64_to_cpu(ja->buckets[i]);
+ j->d[dst].nr = le64_to_cpu(1);
+ }
+ }
+
+ BUG_ON(dst + 1 != nr);
+
+ return 0;
+}
--- /dev/null
+/* SPDX-License-Identifier: GPL-2.0 */
+
+#include "super-io.h"
+#include "vstructs.h"
+
+static inline unsigned bch2_nr_journal_buckets(struct bch_sb_field_journal *j)
+{
+ return j
+ ? (__le64 *) vstruct_end(&j->field) - j->buckets
+ : 0;
+}
+
+static inline unsigned bch2_sb_field_journal_v2_nr_entries(struct bch_sb_field_journal_v2 *j)
+{
+ if (!j)
+ return 0;
+
+ return (struct bch_sb_field_journal_v2_entry *) vstruct_end(&j->field) - &j->d[0];
+}
+
+extern const struct bch_sb_field_ops bch_sb_field_ops_journal;
+extern const struct bch_sb_field_ops bch_sb_field_ops_journal_v2;
+
+int bch2_journal_buckets_to_sb(struct bch_fs *, struct bch_dev *);
#include "io.h"
#include "journal.h"
#include "journal_io.h"
+#include "journal_sb.h"
#include "journal_seq_blacklist.h"
#include "replicas.h"
#include "quota.h"
memcpy(dst->compat, src->compat, sizeof(dst->compat));
for (i = 0; i < BCH_SB_FIELD_NR; i++) {
- if (i == BCH_SB_FIELD_journal)
+ if ((1U << i) & BCH_SINGLE_DEVICE_SB_FIELDS)
continue;
src_f = bch2_sb_field_get(src, i);
mutex_unlock(&c->sb_lock);
}
-/* BCH_SB_FIELD_journal: */
-
-static int u64_cmp(const void *_l, const void *_r)
-{
- u64 l = *((const u64 *) _l), r = *((const u64 *) _r);
-
- return l < r ? -1 : l > r ? 1 : 0;
-}
-
-static int bch2_sb_journal_validate(struct bch_sb *sb,
- struct bch_sb_field *f,
- struct printbuf *err)
-{
- struct bch_sb_field_journal *journal = field_to_type(f, journal);
- struct bch_member *m = bch2_sb_get_members(sb)->members + sb->dev_idx;
- int ret = -EINVAL;
- unsigned nr;
- unsigned i;
- u64 *b;
-
- nr = bch2_nr_journal_buckets(journal);
- if (!nr)
- return 0;
-
- b = kmalloc_array(sizeof(u64), nr, GFP_KERNEL);
- if (!b)
- return -ENOMEM;
-
- for (i = 0; i < nr; i++)
- b[i] = le64_to_cpu(journal->buckets[i]);
-
- sort(b, nr, sizeof(u64), u64_cmp, NULL);
-
- if (!b[0]) {
- pr_buf(err, "journal bucket at sector 0");
- goto err;
- }
-
- if (b[0] < le16_to_cpu(m->first_bucket)) {
- pr_buf(err, "journal bucket %llu before first bucket %u",
- b[0], le16_to_cpu(m->first_bucket));
- goto err;
- }
-
- if (b[nr - 1] >= le64_to_cpu(m->nbuckets)) {
- pr_buf(err, "journal bucket %llu past end of device (nbuckets %llu)",
- b[nr - 1], le64_to_cpu(m->nbuckets));
- goto err;
- }
-
- for (i = 0; i + 1 < nr; i++)
- if (b[i] == b[i + 1]) {
- pr_buf(err, "duplicate journal buckets %llu", b[i]);
- goto err;
- }
-
- ret = 0;
-err:
- kfree(b);
- return ret;
-}
-
-static void bch2_sb_journal_to_text(struct printbuf *out, struct bch_sb *sb,
- struct bch_sb_field *f)
-{
- struct bch_sb_field_journal *journal = field_to_type(f, journal);
- unsigned i, nr = bch2_nr_journal_buckets(journal);
-
- pr_buf(out, "Buckets: ");
- for (i = 0; i < nr; i++)
- pr_buf(out, " %llu", le64_to_cpu(journal->buckets[i]));
- pr_newline(out);
-}
-
-static const struct bch_sb_field_ops bch_sb_field_ops_journal = {
- .validate = bch2_sb_journal_validate,
- .to_text = bch2_sb_journal_to_text,
-};
-
/* BCH_SB_FIELD_members: */
static int bch2_sb_members_validate(struct bch_sb *sb,
__bch2_check_set_feature(c, feat);
}
-/* BCH_SB_FIELD_journal: */
-
-static inline unsigned bch2_nr_journal_buckets(struct bch_sb_field_journal *j)
-{
- return j
- ? (__le64 *) vstruct_end(&j->field) - j->buckets
- : 0;
-}
-
/* BCH_SB_FIELD_members: */
static inline bool bch2_member_exists(struct bch_member *m)
projects / linux-block.git / commitdiff